Predicting the interfacial thermal resistance of electronic packaging materials via machine learning

With the wide application of nanomaterials in electronic devices, the interfacial thermal property has become an indispensable parameter in the thermal management of electronic packaging. Materials informatics, the combination of simulation/experimentation and machine learning, which is now generating a great deal of attention as a tool to accelerating the process of searching for new interfacial thermal materials. In this work, we have proposed a high-throughput screening framework to design interfacial materials with high/low interfacial thermal conductance. The prediction model was trained and established based on the collection of interfacial thermal resistance via the diffuse mismatch model. The input properties for diffuse mismatch model were obtained by the first-principles harmonic lattice dynamics calculation with 207 crystals. The crystal structure parameters and elemental properties are adopted as descriptors as they are quickly accessible from the materials database. The prediction models for interface thermal resistance were successfully built up by using the random forest and support vector machine algorithms. The descriptors of energy, lattice constant, volume and density of the crystals were found to have strong impact on determining the interface thermal resistance of heterogeneous interfaces.